stage2_orm_kto_set2

This model is a fine-tuned version of Qwen/QwQ-32B on the stage2_orm_kto_set2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3478
• Rewards/chosen: 1.9112
• Logps/chosen: -432.7298
• Logits/chosen: -907315200.0
• Kl: 0.0

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-06
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 8
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Logps/chosen	Logits/chosen
0.2781	0.6	75	0.2565	8.4292	-367.5502	-461496661.3333	47.0870
0.096	1.2	150	0.4566	-1.2513	-464.3549	-598909781.3333	0.0
0.1185	1.8	225	0.3182	1.2599	-439.2429	-654441386.6667	0.0
0.012	2.4	300	0.3210	2.5768	-426.0743	-780268800.0	0.0
0.0102	3.0	375	0.3478	1.9112	-432.7298	-907315200.0	0.0

Framework versions

• Transformers 4.55.0
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.21.1